Rehabilitation Setting and Associated Mortality and Medical Stability Among Persons With Amputations

Data Sources 

We studied data from the Center for Medicare & Medicaid Services (CMS, formerly the Health Care Finance Administration). Medicare claims files for 1996 and 1997 were used to examine discharge destination and 12-month outcomes among elderly persons undergoing dysvascular amputations in the United States. These data, which have been widely used in health services research to examine issues ranging from cost of care to quality of care and long-term outcomes,23, 24, 25 represent the health care experiences of Medicare beneficiaries nationwide. Both acute and postacute care services (eg, home health care, skilled nursing) are available from the data. In particular, we used CMS's standard analytic files, which contain records on utilization and payments of all Medicare-covered services, including those covered by Medicare Part A Hospital Insurance (eg, acute care and rehabilitation hospitals, SNFs, home health, hospice) and Medicare Part B Medical Insurance (eg, outpatient and physician/supplier services). These data were supplemented by the Medicare denominator file, which contains basic demographic information about the beneficiary, including enrollment and entitlement indicators, date of birth, sex, race, and date of death. With its large and nationally representative data, the Medicare claims data provided a unique opportunity to examine the postamputation care trajectories for a large sample of elderly persons in the United States. The study was approved by the appropriate institutional review boards.

Study Population and Definitions 

As a first step, an algorithm was developed to select all patients undergoing an amputation-related procedure in 1996 based on the 5% Medicare inpatient standard analytic file.1, 2 Specifically, the initial study population consisted of claims with a procedure code for upper- or lower-limb amputations (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes 84.00–84.09, 84.10–84.19, 84.91). All amputation-related discharges were then classified hierarchically into mutually exclusive categories according to etiology as (1) trauma related (ICD-9-CM codes 810–839, or 880–884, or 885–887.7, or 895–897.7, or 925–929, or 942–949, or 958–959), (2) cancer related (ICD-9-CM codes 170.4–170.8, or 171.2–171.3, or 172.6–172.7), (3) associated with dysvascular diseases (ICD-9-CM codes 040.0, or 250.0–250.9, or 440.0–440.9, or 442.0–442.9, or 443.0–443.9, or 444.0–444.9, or 682.0–682.9, or 686.0–686.9, or 707.0–707.9, or 728.86, or 730.0–730.9, or 785.0–785.9), or (4) other etiology. Records grouped in the residual category of other etiology were then examined in depth. Whenever appropriate, records were reclassified into one of the 3 main groups and our classification algorithm was refined to better reflect the etiology of the amputation. The refinement of our classification algorithm continued until records could no longer be categorized into one of the more specific groups. The final category of other etiology, which contained less than 2% of the sample, included primarily discharges for amputations due to complications of procedures, internal derangement of joints, and other disorders of joints. These unclassified cases and those associated with persons undergoing amputations secondary to trauma or cancer were excluded from the analyses. For the purposes of this study, only persons with lower-limb amputations at the foot, transtibial, and transfemoral levels were included (ie, persons with upper-limb amputations only and those with amputations at the toe level only were excluded from the analysis). For each person identified with a lower-limb “major” dysvascular amputation in 1996, all (1) inpatient, (2) outpatient, (3) home health, (4) SNF, (5) hospice, (6) durable medical equipment (DME), and (7) physician supplier Part B claims were extracted for a 365-day period after the index amputation surgery. Identification of claims was based on each patient's Medicare health insurance claim number. Key data items available for each institutional claim included a principal and up to 10 secondary ICD-9-CM−coded diagnoses and a principal and up to 5 additional procedure codes. Key data items for Part B and DME claims included a principal and up to 5 secondary ICD-9-CM−coded diagnoses.

Care trajectories were examined with a special focus on the disposition immediately after acute surgical care during the hospitalization in which the amputation procedure occurred. Utilizing claims files as described above, we derived the postacute (postamputation) setting and classified them into 3 categories constituting our dependent variables of interest—comprehensive inpatient rehabilitation, SNF, and discharge home. Excluded from this analysis were patients who died during the acute care hospital stay and the small number of amputees who were discharged to hospice care or transferred to other acute care facilities after the amputation.

Each person's index dysvascular amputation was classified into mutually exclusive categories according to the level of the amputation: foot or ankle, transtibial (including below-knee and Syme's amputations), or transfemoral (including through-knee, above-knee, hip disarticulation, and pelvic-level amputations) amputations. Whenever claims for 2 or more amputation procedures were performed during the same hospitalization, the most proximal level of amputation was chosen for classification. In the absence of information regarding laterality of the index amputation, subsequent amputations (ie, amputations identified during the 365-day period post–index amputation discharge) at a higher level were considered to involve the same lower limb as the initial amputation. Subsequent amputations at the same or lower levels than the index amputation were classified as contralateral amputations.

The outcomes of interest for this analysis were mortality, probability and number of reamputations, receipt of a prosthetic device, and number of subsequent hospital admissions. The reference period for all outcomes was 12 months post–index amputation surgery. The number of major comorbidities, including presence of diabetes, was identified by applying the algorithm of Deyo et al26 to all 1996 and 1997 inpatient, outpatient, and Part B files for all study persons.

Data Analysis 

The patterns of reamputation and mortality, along with prosthesis receipt and admissions to a hospital for nonamputation causes, across persons receiving postamputation rehabilitative care in the alternative postacute settings—inpatient rehabilitation, SNF, and home—were contrasted and compared using univariate (t, χ2) test statistics. Nonparametric (Wilcoxon rank-sum, Mann-Whitney U) test statistics were used to examine outcome variations across these different care settings.

Multivariate techniques using a multinomial logit specification were used to examine the independent effects of sociodemographic, geographic, and health and amputation-related characteristics on the likelihood of discharge to alternative settings (home, inpatient rehabilitation, SNF). Because the dependent variable in this analysis was a polychotomous, unordered categoric variable ranging from 0 (discharged home, the reference category) to 2 (discharged to an SNF), conventional linear regression techniques were inappropriate. Our approach, instead, was to analyze discharge destinations in the general framework of probability models.27, 28, 29

Disposition to inpatient rehabilitation was defined as any admission that occurred within 2 days of discharge from the acute care stay in which the index amputation occurred. Both transfers to free-standing rehabilitation hospitals or a rehabilitation unit within the same or different acute care hospital were considered inpatient rehabilitation dispositions. Persons admitted to nursing homes (within that same time frame) and patients admitted to subacute rehabilitation facilities were coded as discharges to SNFs.

A multivariate probabilistic binary choice model (probit specification) was used to examine the effects of discharge setting and other factors on dichotomous outcomes (mortality, any reamputation, receipt of prosthetic device), and a negative binomial specification was used to model count data outcomes (number of reamputations, number of subsequent hospital admissions). P values presented are based on these analyses. To provide a more intuitively appealing presentation of the impact of alternative postacute care settings on outcomes, adjusted probabilities and predicted values were calculated using the parameter estimates from these multivariate models for each outcome. Differences in these adjusted probabilities and outcome values can be interpreted as the marginal effects of the (immediate) postacute care setting in each outcome.

The following sociodemographic and amputation-related variables were included to control for possible confounding of the relationship between discharge disposition and outcomes: patients' ages (<65y, 65–74y, 75–84y, >85y), sex, race (black, Hispanic, white, other), and initial amputation levels (foot or ankle, transtibial, transfemoral) and the region of the United States where patients resided (Midwest, Northeast, South, West). Case-mix adjustment was based on the number of comorbidities algorithm proposed by Deyo et al26 and expanded by Elixhauser et al.30 Major comorbidities were assessed using inpatient, outpatient, and Part B Medicare claims for each study subject during the 12-month period in the year of the index amputation and included indicators for renal diseases, malignancies, myocardial infarction, congestive heart failure, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatologic diseases, hypertension, peptic ulcer disease, mild and severe liver disease, diabetes, diabetes with complications, coagulopathy, acquired immune deficiency syndrome, paralysis, mental disorders (including depression), and drug and alcohol abuse. As applied, the algorithm provides a comprehensive approach to identifying comorbidities to control for factors affecting both disposition to specific postacute rehabilitation settings and outcomes conditional on rehabilitation setting.

All data analyses were conducted using SASa and Statab statistical software. Unless otherwise noted, only differences that are statistically significant at a P level less than .05 are discussed in the text.

Sample Characteristics 

A total of 2468 persons in the 5% random Medicare sample were identified from the claims data as undergoing a major lower-limb amputation in 1996. Table 1 presents the distribution of the sample, by discharge location, according to sociodemographic and amputation-related characteristics. Patients discharged to SNFs were generally older and more likely to be female than patients discharged to either inpatient rehabilitation or home. Although there were no statistically significant differences among patients discharged to alternative settings in terms of number of comorbidities, patients discharged to SNFs were less likely to have diabetes than those discharged to inpatient rehabilitation or home. Differences in discharge destination by level of amputation were also evident from these data. Patients discharged directly to an SNF were more likely to have had an amputation at the highest (transfemoral) level than patients discharged to inpatient rehabilitation (48% vs 34%, respectively). Patients discharged to inpatient rehabilitation, on the other hand, were about 4 times more likely to have a foot or ankle amputation and twice as likely to have an amputation at the transtibial level than those discharged home. Finally, there were also marked differences in the distribution of discharge destination by geographic region.

Table 1. Sample Characteristics

	Characteristics	Overall (N=2468)	By Discharge Destination
			Rehabilitation (n=344)	SNF (n=903)	Home (n=1221)
	Mean age (y)	74.6	71.0⁎†	77.3‡	73.6
	Age at index amputation (y)
	<65	390(16)	74(22)⁎	91(10)‡	225(18)
	65–74	763(31)	121(35)⁎	254(28)	388(32)
	75–84	801(32)	124(36)†	309(34)‡	368(30)
	≥85	514(21)	25(7)⁎†	249(28)‡	240(20)
	Sex
	Male	1216(49)	185(54)⁎	401(44)‡	630(52)
	Race/ethnicity
	White	1640(66)	226(66)	608(67)	806(66)
	Black	668(27)	98(28)	238(26)	332(28)
	Hispanic	73(3)	10(3)	22(3)	41(3)
	Other	87(4)	10(3)	35(4)	42(3)
	Comorbidities
	No comorbidities	36(1.5)	2(0.5)	10(1.1)	24(2)
	1	233(9)	29(8)	78(9)	126(10)
	2–3	1093(43)	158(46)	405(45)	530(43)
	≥4	1106(45)	155(45)	413(46)	538(44)
	Diabetes	1736(70)	259(75)⁎	604(67)‡	873(72)
	Initial amputation level
	Foot or ankle	379(15)	18(5)⁎†	113(13)‡	248(20)
	Transtibial	949(39)	209(61)⁎†	356(39)‡	384(32)
	Transfemoral	1135(46)	117(34)⁎†	432(48)	586(48)
	Census region
	Northeast	503(21)	81(23)	185(20)	237(21)
	Midwest	562(23)	102(30)⁎†	217(24)	243(21)
	South	1055(44)	128(37)†	367(41)‡	560(48)
	West	284(12)	33(10)	132(15)‡	119(10)
	Mortality	1018(41)	77(22)⁎†	355(39)‡	586(48)

NOTE. Values are n or n (%).

⁎ Significant differences between rehabilitation and SNF at the .05 level. † Significant differences between rehabilitation and home at the .05 level. ‡ Significant differences between SNF and home at the .05 level.

Factors Affecting Discharge Destination 

Bivariate statistics shown in table 1 suggest variation by age, sex, presence of diabetes, and initial level of amputation in the rates at which patients were discharged to alternate settings. In particular, the oldest old (≥75y) and female patients were significantly more likely than their younger counterparts and male amputees to be discharged to a nursing home. Compared with those discharged to a nursing home or home, amputees receiving postacute care at inpatient rehabilitation facilities were more likely to have experienced amputations at the transtibial level and less likely to have a foot or ankle or transfemoral amputation. There were also marked differences in discharge destination by geographic region. Amputees in the Midwest were more likely to be discharged to an inpatient rehabilitation facility than to an SNF or home, whereas those in the South were significantly more likely to be discharged home than to any other postacute care setting. Finally, mortality was significantly higher among those discharged home relative to those discharged to any other setting (48% vs 22% and 39%, among those discharged to inpatient rehabilitation or nursing home).

Table 2 presents the relative risk (RR) of discharge to alternative postacute care settings (inpatient rehabilitation and SNF; reference group, home) associated with each of the demographic, health, and amputation-related characteristics included in our multinomial logit regression model. RRs exceeding 1 indicate a positive impact of a given characteristic (eg, male sex) relative to its reference category (eg, female sex) on the likelihood of discharge to a given setting as compared with the likelihood of discharge home (the reference group). Values less than 1 indicate a lower RR of discharge to a given setting compared with discharge home for a given characteristic (relative to its reference category).

Table 2. RR and Confidence Interval of Factors Affecting Discharge Disposition

	Discharge Disposition	Rehabilitation vs Home	SNF vs Home
	Age at first amputation (y)
	<65	Reference group	Reference group
	65–74	0.94(0.66–1.33)	1.63(1.21–2.19)⁎
	75–84	1.02(0.72–1.45)	2.10(1.56–2.83)⁎
	≥85	0.32(0.19–0.53)⁎	2.59(1.87–3.60)⁎
	Sex
	Male	1.01(0.78–1.31)	0.83(0.69–1.00)
	Race or ethnicity
	White	Reference group	Reference group
	Black	1.16(0.87–1.55)	0.97(0.78–1.19)
	Hispanic	0.80(0.38–1.68)	0.75(0.43–1.30)
	Other	0.96(0.46–1.99)	1.21(0.75–1.95)
	Comorbidities	0.97(0.89–1.06)	1.07(1.01–1.14)⁎
	Diabetes	0.98(0.72–1.35)	0.85(0.68–1.06)
	Initial amputation level
	Foot or ankle	Reference group	Reference group
	Transtibial	8.04(4.82–13.41)⁎	1.87(1.42–2.45)⁎
	Transfemoral	3.30(1.94–5.60)⁎	1.33(1.01–1.74)⁎
	Census region
	Northeast	Reference group	Reference group
	Midwest	1.35(0.95–1.92)	1.39(1.06–1.81)⁎
	West	0.77(0.48–1.24)	1.91(1.39–2.63)⁎
	South	0.73(0.53–1.02)	1.03(0.81–1.30)
	Likelihood ratio test	268.7⁎
	Percentage of correct predictions	72

⁎ Significant at the .05 level.

Multivariate results from the statistical analyses that control for a wide array of sociodemographic, case-mix adjusters, and amputation-related characteristics are shown in table 2 and are generally consistent with the patterns described for bivariate associations. Persons undergoing amputations at higher levels—transtibial and transfemoral—were also at higher RR of discharge to inpatient rehabilitation compared with persons with foot amputations (RR=8.0 for amputations at the transtibial levels; RR=3.3 for amputations at the transfemoral level). With the exception of the oldest old (patients aged ≥85y at the time of their amputation), who were less likely to be discharged to an inpatient rehabilitation facility (RR=.32), there were no demographic or regional differences in the probability of discharge to an inpatient rehabilitation (as opposed to being sent home) after the acute care amputation stay.

The RR of discharge to an SNF increased monotonically with age, number of comorbidities, and higher amputation levels (relative to foot). Amputees who were female were also at a significantly higher risk of discharge to an SNF than their male counterparts, although patient race or ethnicity did not influence the RR of discharge to an institutional postacute care setting relative to home. Finally, patients living in the Midwest and those in the West were more likely to be discharged to nursing homes or subacute care facilities than home (RR=1.39, RR=1.91, respectively) than amputees in the Northeast region of the country, even after controls for preexisting comorbid conditions and demographic, health, and amputation-related characteristics.

Postacute Care Setting and Outcomes 

The associations between discharge destination and 12-month mortality, reamputations, prescription of a prosthetic device, and subsequent inpatient admissions are shown in table 3. Adjusted probabilities (presented as percentages) shown in table 3 are based on regression models that control for the same set of patient sociodemographics, comorbid conditions, and amputation characteristics described above.

Table 3. Marginal Effect of Disposition to Alternative Settings on Outcomes

	Disposition	Rehab	SNF	Home	Rehab vs Home, % (P)	SNF vs Home, % (P)	Rehab vs SNF, % (P)
	One-year survival	75%	63%	51%	24⁎(.00)	12⁎(.00)	12⁎(.00)
	Reamputation	18%	23%	25%	−7⁎(.02)	−2(.47)	−5(.10)
	No. of reamputations	0.23	0.29	0.32	−28⁎(.05)	−9(.48)	−20(.15)
	Prosthesis acquisition	73%	58%	49%	24⁎(.00)	9⁎(.00)	15⁎(.00)
	No. of subsequent acute care (nonamputation) admissions	2.09	2.32	2.43	−14⁎(.00)	−4(.25)	−10⁎(.05)

NOTE. Adjusted percentages and outcome values derived from multivariate regression models controlling for patient sociodemographic characteristics, amputation level, and comorbidities.

⁎ Differences significant at the P<.05 level.

Overall, amputees discharged to a comprehensive inpatient rehabilitation facility or unit experienced significantly better outcomes than those discharged to a nursing home, even after controlling for amputation level, comorbidities, and other confounders (see table 3). Specifically, the adjusted probabilities suggest that persons discharged to inpatient rehabilitation experienced a significantly higher survival rate than persons with a similar profile who were discharged to SNFs or home (12% and 24% higher survival probability, respectively). Similarly, those discharged to inpatient rehabilitation were 7% less likely to experience a reamputation and 24% more likely to have been prescribed a prosthetic device than those discharged home. Discharge to inpatient rehabilitation after amputations reduced admissions for other issues besides amputations in the ensuing year. Subsequent acute care admissions were 14% (P<.01) less if discharged to a rehabilitation unit compared with going home and 10% (P<.05) less than if the patient was sent to an SNF (see table 3). Persons receiving immediate postacute care at SNFs also experienced better outcomes than those discharged home. Specifically, those discharged to nursing facilities were 12% more likely to be alive after 12 months after their index amputation; they were also 9% more likely to have received a prosthesis than their counterparts who were discharged directly home.

As shown in table 4, the pattern of better outcomes among persons discharged to inpatient rehabilitation facilities held true for all amputation levels. Despite the relatively small cell sizes for amputations at the foot level, the adjusted probabilities controlling for comorbidities show that only 5% of patients discharged to inpatient rehabilitation experienced a reamputation in contrast to 42% for SNF discharges and 39% for discharges home (both differences significant at P<.05). There were pronounced 87% (P<.05) reductions in the number of subsequent amputations when rehabilitation disposition was compared with both home and SNF discharges (see table 4). Among transtibial amputees, survival and prosthesis prescription were significantly higher among those discharged to inpatient rehabilitation relative to those discharged home or to an SNF (19% and 10% increase in survival relative to home and SNF, respectively; 14% and 8% increase in prosthesis prescription relative to home and SNF, respectively). Finally, among patients undergoing amputation at the highest (transfemoral) level, discharge to a rehabilitation facility resulted in significantly and substantially improved outcomes relative to discharges to either an SNF or directly home across all dimensions considered.

Table 4. Marginal Effect of Disposition to Alternative Settings on Outcomes, by Initial Amputation Level

	Disposition	Rehab (n=18)	SNF (n=113)	Home (n=248)	Rehab vs Home, % (P)	SNF vs Home, % (P)	Rehab vs SNF, % (P)
	Foot
	One-year survival	70%	75%	70%	0(.992)	5(.297)	−5(.622)
	Any reamputation	5%	42%	39%	−34⁎(.013)	3(.645)	−37⁎(.010)
	No. of reamputations	0.07	0.58	0.54	−87⁎(.041)	7(.660)	−87⁎(.034)
	Prosthesis prescription	61%	52%	39%	22(.110)	13⁎(.054)	9(.513)
	No. of subsequent acute (nonamputation) admissions	2.27	2.47	2.75	−17%(.317)	−10%(.247)	−8%(.666)
	Transtibial	Rehab (n=209)	SNF (n=356)	Home (n=384)
	One-year survival	76%	66%	57%	19⁎(.000)	9⁎(.007)	10⁎(.017)
	Any reamputation	24%	21%	24%	0(.855)	−3(.451)	3(.617)
	No. of reamputations	0.30	0.24	0.27	11(.648)	−11(.553)	25(.319)
	Prosthesis prescription	81%	73%	67%	14⁎(.002)	6(.156)	8⁎(.053)
	No. of subsequent acute (nonamputation) admissions	2.11	2.34	2.29	−8(.253)	2(.726)	−10(.149)
	Transfemoral	Rehab (n=117)	SNF (n=432)	Home (n=586)
	One-year survival	74%	56%	40%	34⁎(.000)	16⁎(.000)	18⁎(.000)
	Reamputations	7%	18%	19%	−12⁎(.004)	−1(.694)	−11⁎(.009)
	No. of reamputations	0.07	0.21	0.24	−70⁎(.002)	−13(.502)	−67⁎(.009)
	Prosthesis use	72%	44%	35%	37⁎(.000)	9⁎(.031)	28⁎(.000)
	No. of subsequent acute (nonamputation) admissions	1.84	2.22	2.40	−23⁎(.004)	−8(.229)	−17⁎(.046)

NOTE. Adjusted percentages and outcome values derived from interaction terms (amputation level and discharge setting) in multivariate regression models controlling for patient sociodemographic characteristics, amputation level, and comorbidities.

⁎ P significant at the .05 level.

Study Limitations 

Although this study provided important insights into a number of relevant outcomes for dysvascular amputees managed in different care settings, the use of claims data to estimate the impact of postacute care setting on outcomes has inherent limitations that merit comment. First, information on other outcomes of interest, including physical and social functioning, QOL, pain, and depression, are not available from administrative data. Such outcomes must be examined through clinical studies designed to assess these measures through primary data collection. In addition, clinical information on severity of disease, complications secondary to the amputation, and the patient's underlying health status and physical functioning—factors that might affect discharge disposition as well as outcomes conditional on postacute care setting—is generally limited in administrative data. Although Medicare claims data for services received during the year of the index amputation provided us with powerful indicators of preexisting health and allowed us to more accurately adjust for differences in health status across patients discharged to alternative settings in our analysis of long-term outcomes, no information on their severity or the extent to which they might impair function is available from these data.

A standard concern that arises in studies examining the relationship between discharge destination and outcomes is the potentially biasing effects of nonrandom assignment of patients to alternative postacute care settings. Such a bias would occur if, for example, there are unobserved or unmeasured health status issues, which might affect simultaneously discharge destination and outcomes conditional on postacute care setting. One strategy to deal with this problem would be to use a 2-stage procedure in which the discharge destination process is modeled in stage 1 and the outcomes are modeled in stage 2. This instrumental variables method requires that additional variables be included in stage 1 that are predictive of patient destination discharge but that do not affect the second stage outcomes conditional on where the patient received postacute care. Indicators of a patient's social support and access to informal care at home, for example, might have been useful instruments for this purpose but were not available in the Medicare claims data. There are, however, 2 critical issues raised by this approach. First, effects measured through this approach pertain to those marginal patients whose disposition is in fact influenced by the instrumental variables, but this effect may not generalize to the entire population of patients.25 Second, testing the validity of the assumption that instrumental variables have no direct effects on our dependent variables is problematic when the study population is characterized by heterogeneous responses.27 To address potential problems of sample bias, we attempted to control for case mix by including an extensive list of comorbidities, assessed over a 12-month period, as covariates in our regression models. We believe this strategy has definitely minimized any potential sample bias in our study, but it may not have completely eliminated the problem.34, 35

Postamputation discharge evaluation is a very complex decision-making process requiring consideration and reconsideration of an almost overwhelming number of variables from a wide variety of data sources. In addition to medical and surgical status of the patient, the provider must take into consideration patient preferences and constraints, including the degree of family support, ability to pay, ability to participate and benefit from more intense rehabilitation, and desire to go home or remain in the hospital. Our study evaluates the realization of these decisions and controls for only a limited number of potential confounders and thus certainly does not provide a complete picture of how and why providers and patients make decisions regarding postacute settings. Although we cannot completely rule out the possibility that the outcomes were simply a reflection of appropriate assignment of patients to the correct rehabilitation setting, this is unlikely to be the driving force behind the results, given the marked differences in outcomes given the statistical equalization of patients across the 3 discharge settings and the high proportion (72%) of correct predictions of the discharge destination model using the available set of control variables. Nonetheless, a better understanding of the nature of postacute care decisions remains a necessary goal for future studies.

Despite these qualifications, we believe that documentation of compelling evidence linking comprehensive inpatient rehabilitation services to: (1) improved 1-year survival rates, (2) reduced subsequent amputations, (3) increased prescription for a prosthesis, and (4) improved medical stability through reduced hospital admissions is of great interest and should catalyze further study into the underlying process whereby rehabilitation provided in different settings affects outcomes. The relatively large and nationally representative sample used in the study, the objective nature of the outcomes evaluated, and the strength of the statistical inferences make these findings relevant and important from a health policy perspective. Although limited by a relatively smaller set of control variables, the benefits of using such large administrative data sets are great, because they provide generalizable information on the association between care setting and outcomes that would be prohibitively costly to ascertain through randomized controlled trials or primary data collections. Such large-scale health services research studies addressing topics in rehabilitation are rare, limiting the ability of our field to fully assess rehabilitation care delivery in the United States. The outcomes examined here—survival, reamputations, and hospital readmissions—are unambiguous and consistently measured using administrative data, making them ideal for such health services investigations. These outcomes are meaningful from the perspectives of patients, providers, payers, and policymakers alike. The improvements in survival for the entire group and for persons with transtibial and transfemoral amputation levels was perhaps the most compelling positive outcome associated with direct discharge to inpatient rehabilitation, even after controlling for comorbid conditions and other factors likely to influence mortality.